univi.evaluation.denoise_adata#
- univi.evaluation.denoise_adata(model, adata, *, modality, device='cpu', out_layer='denoised_fused', overwrite_X=False, batch_size=512, adata_by_mod=None, layer_by_mod=None, X_key_by_mod=None, use_mean=True, decoder_prefer='auto', decoder_link=None)[source]#
Denoise (reconstruct) a modality.
If adata_by_mod is None: encodes only this modality, decodes this modality (self denoise).
If adata_by_mod is provided: encodes a fused posterior from adata_by_mod (paired/multi-observed), then decodes modality and writes into adata.layers[out_layer].
Returns the modified adata (for chaining).